Dear Experts,
I’d be grateful if anyone could clarify a conceptual point around source reconstruction in SPM. I expect I will be able to have this answered at the forthcoming SPM EEG workshop, but I’d like to be a little further ahead when I arrive there. If answers to these questions are present in the SPM manual or in Litvak et al (2011), I’m afraid I have missed them.
My question is whether the nifti images that I generate after group inversion, and which are used for statistical testing over subjects are constrained to hold only positive values.
I am using 45 files: each is a subject, each file contains four epochs which are averages over trials for four conditions. When running group inversion, I specify the frequency band as the default zero since according to Litvak et al and the SPM manual this means polarity will be included in the time window averaging (i.e., signed values are used). This is because I would like to retain information on the polarity of the source.
The maximum intensity projection plots that follow group inversion do indeed show values that can be positive or negative across the inversion window, variable over subjects. I am assuming that intensity here is something like current, though that may be wrong.
At the following windowing stage, the glass brain is titled “energy evoked”. Taken at face value I would assume this means that intensities have been squared and have lost their sign.
This is consistent with two subsequent observations. When I create a nifti image at the next step in the source reconstruction window, I find when I move crosshairs around the resulting image that all intensities on the map are positive at all points (Two different “Y” intensities can be displayed, the first appears unrelated to the previous glass brain intensities and is non-zero on the scalp and outside the cortex, but the intensity displayed below this is related to the MIP glass brains and is zero outside the cortex).
Second, if I carry these nifti files through to second level stats and run a one sample t-test, I find that pretty much the whole cortex shows significant activation when specifying a contrast of 1 (asking, I believe, the question where are intensities greater than 0 at the group level). Nothing is significant when specifying a contrast of -1 even if alpha is set at an absurd level of 0.80. This is the behaviour I would expect if all individual subject x source intensities are constrained to be no less than zero.
When I run a paired t-test or F test on these nifti images at to examine condition effects, I get more sensible results though extremely weak, requiring there to be no FWE correction. The strong condition effects seen in the scalp ERP (based on signed values) are absent.
Hence my question: is removing sign a necessary step in the source reconstruction method used in SPM and if so, why is it only introduced at the point of windowing for an image?
And my associated questions
Is it essentially wrongheaded to attempt to retain the sign of the current at a source and perform statistical tests over subjects on scores that are positively and negatively signed? If so, this contrasts greatly with the group level analysis of scalp ERPs based on grand averages (i.e., arithmetic means of signed values) and associated variance.
And finally: should source reconstruction thus not be performed on ERP difference waves (i.e. condition effects)
Many thanks for your thoughts,
Tom Sambrook
|